February 27, 2011, [MD]
This week, I am participating in the first Learning Analytics conference in Banff, Alberta. I'm interested in this topic both for my PhD research, and it's also something that P2PU has been very interested in pursuing. I started taking notes in Etherpad, and invited others to join me - most of the notes below were taken by me, but a number of others also contributed (thanks to all of you)!. These notes are very raw, but you might find some of it useful. I've also focused more on the things that are interesting to me, rather than trying to provide a comprehensive overview. I will keep taking notes for the next two days of the conference (as long as I keep having good access to a power bar, my aging MacBook has a battery life of about 1,5 hours by now).
Notes below the fold.
The intelligent curriculum – personalizing experience for individual students – adaptive curriculum
TEKRI – has very deep data about student interaction
Q: Who is going to succeed, who will fail? Not percentage-wise, individuals!
“Who talks to whom, about what, and to what effect” (sociology)
“Who’s connecting, what are they talking about, what’s the impact in the long run?”
Analyzing emails, who should be connected – business analysis – IBM
Marissa Meyer – the physics of data
function of scale
map reduce, hadoop
digital data – data exhaust, trails
learning analytics – measurement, collection, analysis, reporting of data about learners and their contexts, or purposes of understanding and optimizing learning, and the environments in which it occurs
knowledge analytics – text / data mining, info retrieval machine learning, linked data) for processing data to provide representations in forms of which conclusions can be drawn in an automated and domain-aware way
model: learners data + intelligent curriculum > profile, analysis > prediction > personalization, adaptation, intervention
ontology – basic structure or armature around which knowledge base can be built
semantic web cannot exist without social web – cannot be easily produced.
DBpedia as ontology
BBC – semantic data – BBC music, leveraging data from MusicBrainz, DBpedia
authoring + reusability > packaging > learning and collaboration (community, peer-review, presenting, community, learners)
learning context ontology – content structure ontology, content type ontology, user model ontology, learning design ontology, domain ontology (DBpedia)
(IEEE International Conference on Advanced Learning Technologies – “Learning Object Context on the Semantic Web”)
sparlq – language for data mashups – can pull in data from many sources
ubiqitous learning analytics – from lot’s of platforms, formats – aggregates and integrates cannot exist without advanced learning analytics
Ideas from lunch:
****use latent semantic analysis to go through all published papers by all professors at a given university – find interesting linkages (topics, methods, common citations) of professors in different departments – sugget that they get to know each other. “you guys should talk”
FYI (from Joe): LSA might not be the best method to use here, I generally prefer “Concept Forest” (but need to make some experiments!).
visualizing activities for self-reflection and awareness
-learning resource recommendation
research on recommendation for learning
LAK11 pre-course data: http://ariadne.cs.kuleuven.be/monitorwidget-lak11
login: choose any users name
views average time spent of all users, and can highlight individual users (how do they calculate time spent? which resources were they able to suck in? ability to use with P2PU data?) is it OSS?
how does this modify student activity?
privacy issues for universities – this is the nice thing about LAK and other MOOCs
can anonymize – see the lines, but not who the students are
George: How real-time is the data? Currently pre-loaded, we can update every day, but not on the fly. Performance issues.
Time – estimates. Track all interactions.
“Altruistic learning situations vs competitive learning situations (grades are curved, etc)”
Participation metrics might be more meaningful in altruistic learning situations – my example from MOOC: I signed up, but was only active during the first week. There is no penalty to dropping out – so people leave very easily. Presumably I would have learnt a lot more if I had been actively engaged with the course over the six weeks.
Compare to a competitive learning situations, where there might be a participation mark to make sure students generate a number of posts per week – in this case, pure participation statistics would be a lot less meaningful, and you’d need to look much more closely at LSA, discourse analysis to see if they are really doing deep thinking, or just “writing”… (This will be useful in open learning situations too – but we need to get the engagement and participation first. Plus it’s harder, because people have more leeway to define their own learning outcomes… Maybe link to individual learning plans etc?)
Contextualized Attention Metadata
deployed in ROLE-PLE, RWTH-Aachen engineering, Moodle
Track every kind of data such as user clicks
http://bit.ly/laksurv – survey about tool/ LAK11
“Happy to load your data into the tool.” Again: how does this work with distributed learning – this is the holy grail (funny how the MOOC people are so anti-LMS, but yet it’s an LMS that enables this to happen) – (gs: good point. Have you looked at social media monitoring tools? They are starting to address distributed interactions: see: http://www.diigo.com/user/gsiemens/monitoring
sh: @gs thanks! – this will be a core focus of the new P2PU platform – having a solid user-friendly core, which enables us to pull in statistics and info from lot’s of outside platforms, wikis, twitter, blogs, Wikipedia, github – for people to view activity, learners to create portfolios, assessment, and data analysis. (Hoping to make data-dumps of all learner activity available to all researchers on a regular basis / real-time)
We used to think that we could just use tags, and suck in #lak11 stuff from lot’s of platforms into a widget. Two problems: spam (all my blog posts get reposted on lot’s of splogs, with the tags intact, and popular tags are often spammes on Twitter etc), and connecting a learner identity in the LMS to an external contribution (learners might not use same login across different systems) – how to do this easily? Would be interesting to come up with a standard data format for sharing “learning interaction logs”, so that visualizers etc could plug directly into data from P2PU / MOOCs, or even BlackBoard… Do any examples exist?
September RecSysTel 2010: http://adenu.ia.uned.es/workshops/recsystel2010/
dataTel: bit.ly/ieqm (too fast)
request for data sets, they already have datasets, Mendeley (2 million users), APOSDLE, ReMashed, .edunet, Mace, Melt
ROLE: Responsive Open Learner Environment
can suggest new tools of resources on the fly
tools that can support learner self-assessment
The Data Shop – A data analysis service for the learning science community http://pslcdatashop.web.cmu.edu #lak11
How can non-programmers do these kinds of things – lowering barriers
Data is a dish best served raw
Acquisition as data
APIs – offered by many social networks
(currently P2PU doesn’t offer any API – wonder what kinds of APIs would be useful to others, how to design them, etc)
screenscraping – extracting data from loaded webpages straight from html
scraperwiki – tool for this – http://scraperwiki.com/
import to HTML in Google Spreadsheet (grab table, or list?) identify table by the number of the table on the page (or number of list) – http://googlesystem.blogspot.com/2007/09/google-spreadsheets-lets-you-import.html
Data is notoriously unreliable – different dataformats, spelling mistakes, correcting errors (outliers might be valid though)
Hirst answered my question now about transformation. A possible tool is Google Refine.
Google Refine – load in a CSV, find strings that look similar and should be similar http://code.google.com/p/google-refine/
Stanford Data Wrangler: http://vis.stanford.edu/wrangler/
Alfred Essa adds: transformation (one structure to another)
– we can stop trends and variations visually much better than in raw tables
generate Google “heatmap”
Yahoo Pipes ( http://pipes.yahoo.com ) useful for generating KML format used by GoogleMaps and GoogleEarth
The really neat thing is that there is a tool to convert Yahoo Pipes to Python, which means that if Yahoo ever pulls the plug (not impossible, given how many services Yahoo shuts down, you won’t loose your tools): http://blog.ouseful.info/2010/09/30/yahoo-pipes-code-generator/
Scraping “In Our Times”, finding members of the OU (not in a linked data form, neither list of academics in OU). lot’s of work to make things join up.
Gephi is an interactive visualization and exploration platform for all kinds of networks and complex systems http://gephi.org/ <- looks very cool
Use common usernames on different social platforms – not perfect, what if the name is taken on a service, or if someone takes your common username and use it to open an account at a new service. But maybe “good enough” for now. Would be great if there was a unique person identifier, which you could put into a field on all social services (and where you could do reverse lookup) – I guess this tends to be e-mail. Does Twitter have an API, give me the Twitter stream of email@example.com?
if I wanted to follow “best practices” when writing about OU Academics, how would I “link” this data? Perhaps hyperlink their names to their faculty pages? What is the “unique identifier”? (Having personal URLs is great, but not sustainable – your own DNS, maybe you forget to renew it – your institution, maybe you change jobs)…
There are more common login services now that may help. For example, facebook logins.
I wonder how these help coordinate data – is there a way for me to find all the accounts that a Facebook user has logged into? Privacy issues? Also, Facebook isn’t the most generous at giving data out (they love “sucking it in” to Facebook though) Good point.
Create Google Custom Search Engine over just the followers in a certain area. “Roll your own” social search engine. Same with creating a “blog roll” in a certain topic.
I wonder if you can extract semantic relationships via twitter or delicous data?
Academic publishing Funny how much easier it is to create networks of people who retweet each other, or follow each other, than to create a graph of people who co-author papers, or are cited by the same papers… We need unique author IDs, and unique paper IDs. And these need to be used every time a citation / author is mentioned in a paper. There are experiments in both categories, but nothing comprehensive yet.
Not related to anything Interesting to think about the ideal design of such a collaborative note taking tool… Ways of keeping an emerging structure, and inserting notes at appropriate locations. (I’d almost like several small windows on the screen for different kinds of notes – one for speaker notes, one for books to look up, one for process notes etc). Ways of graphing visually (like @sbskmi does with Compendium, live meeting notes)… ways of going through at the end and collate everything – with ppts, notes, tweets etc…
Systematic adoption of learning analytics
When looking at analytics, you have to look at past, present and future trends
Past: What happened?
Present: What is happening now?
Future: What will happen?
Accessible high quality Data
Tom Davenport books on analytics http://www.tomdavenport.com
Have a look at Doug Clow’s conference notes: http://dougclow.wordpress.com/2011/02/27/lak11-learning-analytics-and-knowledge-banff/ Oh very nice find! Thanks to Doug!
Stian – hope it’s ok to mention this here as well: http://www.learninganalytics.net/?p=122 cfp learning and knowledge analytics in ETS (delete if it isn’t)
Theory-based learning analytics (Simon Buckingham-Shum)
learning analytics and sensemaking – important and different from normal chat and social network analysis… google analytics etc designed by people who don’t know much about education
New learning theory – that wasn’t possible before we had learning analytics (ref. big science)
KMI – similar to AU’s TEKRI http://projects.kmi.open.ac.uk/hyperdiscourse/
(Note to all: Knowledge Cartography is brilliant, available here: http://www.springerlink.com/content/978-1-84800-148-0 for people who have access. Dan Suthers, who wrote a chapter, will be here too)
theory-based analytics: assumptions, evidence-based findings, statistical models, instructional models, as well as more academic "theories"theory-based analytics: assumptions, evidence-based findings, statistical models, instructional models, as well as more academic “theories”
The question is whether this has INTEGRITY as a meaningful indicator, and WHO/WHAT ACTS on this data
Another summary/notes of LAK11: http://alfredessa.com/?page\_id=617
A theory might predict future patterns based on analytics.
RAISEonline: learning analytics in English schools
UK higher ed system: analytically “oppressed”
Previous OU study data are best predictors of future success
“operationalize” a definition of “at risk” – complex intersection of well-documented factors that threaten completion rates.
this is a theory of learning – model, empirically based findings, we can make predictions. (model of learning or of engagement?)
Sensemaking (Karl Weick): comprehending, patterning, interaction in pursuit of mutual understanding
Collective intelligence tools and analytics should be designed specifically to minimise breakdowns in sensemaking
Experts might fail to recognize a novel phenomenon… SO tools should pay particular attention to edge-cases, or scaffold critical debates between contrasting issues
Complex systems only seem to make sense retrospectively
Use narrative theory to detect and analyse knowledge-sharing/interpretive stories
Coherent pathways through the data ocean are important
Much knowledge is tacit, shared through discourse, no formal cofidications. Trust is key to flexible sensemaking when the environment changes.
SO scaffold the formation and maintenance of quality learning environments.
Many small signals can build over time and become a significant force/change
SO highlight important events and connections → aggregation and emergent patterns
set of generic dispositions and skills that characterize good learners – if we can teach them a language for this, they can learn to become better learners. deal with ambiguity, ask good questions, etc.
7 dimensions of learning power : http://www.vitalhub.net/index.php?id=8
Changing and Learning
Effective Lifelong Learning Inventory (Ruth Deakin Crick)
When making a blog post – tag it with the relevant indicator, dashboards for teachers.
(Idea: link this with work on badges at P2PU)
knowledge is made more publicly accountable and reasoning is more visible in the talk (ref to research papers).
use Elluminate to analyze exploratory talk. separate learning from social interaction. Taking Mercer’s framework for Exploratory talk – find “canned phrases” that signal the presence of these, run these on the transcript of the text chat. Preliminary evidence: yes, it works.
Xerox Incremental Parser: sophisticated text parser technology. Run on scientific paper – can pick out categories of knowledge level claims.
…the previously proposed…
…is universally accepted…
Integrate this with Cohere human annotation interface. Webcast about this (possibly at URL above)
What comes after threaded-discussion forum?
Analyst-defined visual connection language
Reflective deliberation platform (others being developed at MIT). Similar to Scardamalia & Bereiter (http://www.ikit.org/kbe.html)
Check this chapter out – very good overview of different ways of designing deliberative platforms for deeper learning: http://lilt.ics.hawaii.edu/papers/2008/suthers-2008-cartography.pdf
Videos from ODET 2010: Online Deliberation Emerging Tools – via @sbskmi
dashboards must be learner-facing, learners should know everything about themselves that institutions know about them.
If people know what your metrics are, they will game them – like content farms.Stian Håklev February 27, 2011 Toronto, Canada comments powered by Disqus